AI Music Revolution

When the Suno Generation Isn't the Finish Line

• Josh • Episode 28

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 17:11

Send us Fan Mail

Lots of people treat a Suno generation as the finish line. In this episode, Josh walks through a track he started in Suno and rebuilt almost entirely by hand: the tempo-drift problem that breaks a rebuild before it starts, why the stems are scaffolding you eventually delete, how a constraint became the song's identity, and why clarity, not volume, is what makes a mix feel big. AI moved the starting line. It didn't finish the record.

The Unlock System is JG BeatsLab's methodology for serious musicians working with AI tools. Lane 2 work: human-authored, AI-assisted music creation.

Visit JG BeatsLab: https://www.jgbeatslab.com

Every Thursday morning, I send out a newsletter. One email a week. It's where I share what I'm working on in the studio, the research running in the lab, and the methodology I'm refining in real time. New books, new Blueprints, new findings from the Red Lab Protocol research. If that sounds useful, you can sign up at jgbeatslab.com/newsletter. Thanks for listening, and I'll see you next Friday.

AI Music Revolution is produced by JG BeatsLab LLC, an AI music education company building the methodology, research, and community for serious creators working in Lane 2.

Get more from JG BeatsLab LLC:

Connect:

Contact: josh@jgbeatslab.com

Stop gambling. Start directing.

SPEAKER_00

Hello and welcome to the AI Music Revolution. I am Josh Galilean, the founder of JZ Beats Lab. Today I want to walk you through something I just lived. Because I think it's honestly the single most important shift you can make in how you actually work with AI music tools. And it's something that I think we need to really talk about honestly. So here's a setup. I finished a track recently, as you may or may not know. I'm in a couple bands, I have side music projects and whatnot. Well, I just finished a track recently for one of those side projects. It started out actually as a Suno generation. And by the time it was done, it was something I had rebuilt almost entirely by hand in a DAW. I had new drum stems throughout. I had new percussion elements that I created in there as well. I actually used a lot of log drum sounds, which was really cool. I programmed in new bass, new bass sounds throughout. I redid all of the vocals. I have about a dozen vocal tracks in there. I did a lot of atmospheric additions, a lot of some synthesizer work, just really some sound effects. I just put a lot of creativity into it. And by the time it was all said and done, almost none of the original, actually, none of the original Suno audio was in the mix. So the song was still shaped by what Suno gave me, but it was not made of it. And that I think is the distinction. That's the difference between a track shaped by AI and a track made of AI. And that's the whole thing that I want to talk about today. Because most people who use Suno treat a Suno generation as the finish line. You type in a prompt, you run it, you get something that sounds pretty good on the first few listens, and you're done. You upload it, and I understand that appeal. It's quick. You can really get that dopamine hit pretty quick. And the output can generally sound impressive right out of the gate. But there's a wide gap between an impressive generation and a Finnish record. And that that gap is where the actual work lives. And it's also where the people building real catalogs separate themselves from the people who are just collecting credit-wasted stories. So let me tell you what the process actually looked like for me. Some of the problems I ran into, because the problems to me are what are the most useful part of all of this. So the first thing I had to get right was my own head. I went into the generation with a clear job for the AI. I wasn't asking it for a finished recording. I was asking it for basically a map. Where do the sections land? How does the energy build throughout the song? What's the vocal cadence? Where is the rhythm sitting in this song? That's it. I wanted a sketch, not a master. And that one decision changes everything downstream because the moment you decide the generation is a sketch, you stop trying to fix it and you start trying to learn from it. So when the first output had problems, and it did, it wasn't a disappointment, it was just a reference. It was doing exactly the job that I gave it. And the first real problem showed up pretty fast. So when I pulled the audio into my DAW, I use Reaper, to start building the song out, the timing wouldn't lock to a grid. The track was hovering somewhere around 134, 135 beats per minute, but it was drifting the whole way through. Now, for a lot of workflows, that's a minor annoyance. You'd never really even notice. But for what I was about to do, it was a serious obstacle because I was planning to rebuild this thing with programmed drums, layered percussions, the sound effects, and new bass lines, and all of that needs a very stable grid to sit on. So if my reference audio is drifting against the DAW's tempo, then every single element I place has to be hand-nudged to follow that drift. And that's hours of fighting the AI's imperfections instead of building something clean. The instinct that most people have here is to tempo map the DAW to follow the audio. I actually went the other way. I took the track back into Suno Studio first, and I used the time stretch and quantized tools to lock the whole thing in to a fixed grid before it even entered my DAW. Once it was conformed to a single steady tempo, it became a really stable reference that I could actually build against. And that's a lesson worth holding on to. So let me say it plainly. The AI sketch can be musically useful and technically unusable at the very same time. And those are not contradictions. Sometimes the idea is good and the infrastructure under it is broken, and you have to repair the infrastructure before the idea is even workable. And most people never even diagnose that, they just feel like the track is fighting them and they don't know why. Okay, so now at this point in the process, I have a stable reference. So let's get into the actual rebuild. I started with the drums because the song lived on rhythm, and honestly, because you know, drums are where I'm most comfortable making detailed decisions being a drummer. So then I added percussions. I built up as stack layers from one shots and different sound effects rather than one single kit. I really went creative on that. And then I added bass. And the whole time I was doing this, the Suno stems were doing exactly one job. They were telling me where things go, where the section enters, how hard the chorus hits, how how long things run. They were the blueprint taped to the wall, if you will. And the moment my rebuilt parts could carry the song on their own, I deleted the Sunos stems out of the mix entirely. And that's the part that I want to make sure you don't skip. It's the part that matters the most because there's there's a weak version of this workflow where you take the AI stems that you just and you just and you just process them. You just EQ them, polish them until they sound you know a little bit better. And then there's a strong version where you use the stems to understand the song, and then you build the song without them. That second one is the entire difference between remixing AI output and actually producing your own record from an AI starting point. One of those leaves you with a slightly cleaned up version of what everybody else has. The other leaves you with something that is truly, truly yours. Now I want to talk about a decision that did more for this track than anything else because it ties into something bigger than just this one song. So early in the process, I made the call to build this track without a particular instrument that would normally expect to carry a song like this. And I want to be honest with you about why. It started as a limitation, not an artistic statement. That instrument programmed never sounds right to me. So instead of faking something I didn't want and couldn't get to sound good, I just cut it out entirely and I handed its job to other elements. Distorted bass, mechanical percussion, sound sound design. And here's what happened: the the constraint forced a more original result. Nothing on the track is pretending to be a normal band in a normal room because it can't. Because I removed the piece that would have led it. The limitation stopped being a limitation. The exact moment I treated it as the identity of the song instead of a hole I needed to fill. And that happens way, way, way more often than people expect. The thing you can't do well will sometimes push you towards something that only you would make. So before you go chasing a tool or a plug-in to fill in the thing you're missing, ask whether the thing you're missing is actually pointing you at your own sound. And by the way, in case you were wondering, this particular instrument was the guitar. Alright, now let's get into the mix because there's a specific lesson here that I think about all the time. When the arrangement finally came together, the track had it had real mass, it was big, but the lead vocal was getting buried. Everything was fighting it. The percussion, the distorted bass, the atmosphere, all of it was kind of crowding around the same space as the voice. And the obvious move, the one most people reach for, is to just turn the vocal up. But but turning the vocal up rarely fixes that. It just makes a loud thing louder inside of an already crowded mix. You don't get clarity, you get a louder traffic jam. So instead, I use the vocal itself as a trigger. And I had the competing elements ducked down just slightly every time the vocal came through. So we're talking small amounts, one to three decibels at tops. Nothing that you could ever really consciously hear as pumping in the song. The listener should never notice the ducking is even happening. They should just notice that, hey, the vocal suddenly cuts through very cleanly right there. And here's a result that honestly it surprised me, and it surprises me every single time I do it. The mix actually got bigger, not smaller. Bigger because when the elements stop masking each other, when they stop fighting for the same frequency space, the whole thing actually feels larger, even though nothing actually got louder. The clarity, clarity reads as as size. Sit with that for a moment because it's counterintuitive, and it's the one, it's one of the most useful things that I know about mixing. When something feels small or cluttered, the the answer is usually not more volume, it's actually less masking. And then the last decision, mastering, which is really a lesson about knowing when to stop. The final master landed at a loudness that was already right where it needed to be for streaming, with the peak sitting close to the ceiling. And the temptation at that stage is always to push it a little harder, make it a little louder. I play a lot of heavier music, and so loud, loud, loud, squeeze more out of it. And I didn't do that here because a track did not need to be louder, and pushing it harder would have flattened the drums and actually made the whole thing feel smaller. So the only thing I did at the end was a safety pass to protect the peaks for distribution. That's it. The job wasn't at the end, it wasn't loudness, it was preservation. Make it safe, not loud. Knowing when to stop is actually a skill and has won the loudness wars, trained a whole generation of people right out of. So let me bring this all the way back around because I don't want any anyone to hear this as me banishing AI tools. I'm not. I use them, I love them. It's it's this particular situation, it saved me real time at the front of the process, and it surfaced production decisions I might not have found as fast on my own. And that's that's real value, and I'm not going to pretend otherwise. But here's what it did not do. It did not write the lyrics, it did not fix the timing, it did not rebuild the drums, it did not design the sound, it did not mix the vocal, and it did not make a single judgment call about what this song was supposed to be. AI moved the starting line, it did not finish the record, and that's that's the whole game right there. If you treat the generation as a sketch and then you do the work after it, you end up with something that is genuinely yours. If you treat the generation as the finish line, you end up with exactly what everybody else has because everybody else is using the same tools the same lazy way. So that's my challenge to you this week. Next time you get a generation you're excited about, don't upload it. Ask yourself what it's actually a sketch of, and then go and build that thing. Use AI to start, use your craft to finish. That's it for this one. I'm Josh Gellalan. This has been the AI Music Revolution, and I will talk to you in the next one.